摘要 |
PURPOSE:To efficiently acquire semantic knowledge concerning a natural language by calculating distance scale as the conceptional similarity of words in the analysis and generation processing of the natural language. CONSTITUTION:This device is provided with a matrix transformer part 11 for transforming the value of an element to the three kinds of values such as 'true,' 'flase' and 'unknown' from the relation matrix of normalized words provided from the co-occurence relation or the like of words in a sentence data base, inter-word conceptional distance calculation part 12 for calculating the distance between the words from the transformed relation matrix, relation matrix recalculation part 13 for calculating the new relation matrix by utilizing the inter-word distance calculated at the part 12, and the recalculation part 13 is provided with an interactive relation matrix correction part 14 for manually confirming the calculated result concerning the truth/false of relation is doubtful in order to cancel the non-certainness of the calculated result. |